1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * UDP over IPv6 4 * Linux INET6 implementation 5 * 6 * Authors: 7 * Pedro Roque <roque@di.fc.ul.pt> 8 * 9 * Based on linux/ipv4/udp.c 10 * 11 * Fixes: 12 * Hideaki YOSHIFUJI : sin6_scope_id support 13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind 15 * a single port at the same time. 16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data 17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file. 18 */ 19 20 #include <linux/bpf-cgroup.h> 21 #include <linux/errno.h> 22 #include <linux/types.h> 23 #include <linux/socket.h> 24 #include <linux/sockios.h> 25 #include <linux/net.h> 26 #include <linux/in6.h> 27 #include <linux/netdevice.h> 28 #include <linux/if_arp.h> 29 #include <linux/ipv6.h> 30 #include <linux/icmpv6.h> 31 #include <linux/init.h> 32 #include <linux/module.h> 33 #include <linux/skbuff.h> 34 #include <linux/slab.h> 35 #include <linux/uaccess.h> 36 #include <linux/indirect_call_wrapper.h> 37 #include <trace/events/udp.h> 38 39 #include <net/addrconf.h> 40 #include <net/ndisc.h> 41 #include <net/protocol.h> 42 #include <net/transp_v6.h> 43 #include <net/ip6_route.h> 44 #include <net/raw.h> 45 #include <net/seg6.h> 46 #include <net/tcp_states.h> 47 #include <net/ip6_checksum.h> 48 #include <net/ip6_tunnel.h> 49 #include <trace/events/udp.h> 50 #include <net/xfrm.h> 51 #include <net/inet_hashtables.h> 52 #include <net/inet6_hashtables.h> 53 #include <net/busy_poll.h> 54 #include <net/sock_reuseport.h> 55 #include <net/gro.h> 56 57 #include <linux/proc_fs.h> 58 #include <linux/seq_file.h> 59 #include <trace/events/skb.h> 60 #include "udp_impl.h" 61 62 static void udpv6_destruct_sock(struct sock *sk) 63 { 64 udp_destruct_common(sk); 65 inet6_sock_destruct(sk); 66 } 67 68 int udpv6_init_sock(struct sock *sk) 69 { 70 udp_lib_init_sock(sk); 71 sk->sk_destruct = udpv6_destruct_sock; 72 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 73 return 0; 74 } 75 76 INDIRECT_CALLABLE_SCOPE 77 u32 udp6_ehashfn(const struct net *net, 78 const struct in6_addr *laddr, 79 const u16 lport, 80 const struct in6_addr *faddr, 81 const __be16 fport) 82 { 83 u32 lhash, fhash; 84 85 net_get_random_once(&udp6_ehash_secret, 86 sizeof(udp6_ehash_secret)); 87 net_get_random_once(&udp_ipv6_hash_secret, 88 sizeof(udp_ipv6_hash_secret)); 89 90 lhash = (__force u32)laddr->s6_addr32[3]; 91 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); 92 93 return __inet6_ehashfn(lhash, lport, fhash, fport, 94 udp6_ehash_secret + net_hash_mix(net)); 95 } 96 97 int udp_v6_get_port(struct sock *sk, unsigned short snum) 98 { 99 unsigned int hash2_nulladdr = 100 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum); 101 unsigned int hash2_partial = 102 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); 103 104 /* precompute partial secondary hash */ 105 udp_sk(sk)->udp_portaddr_hash = hash2_partial; 106 return udp_lib_get_port(sk, snum, hash2_nulladdr); 107 } 108 109 void udp_v6_rehash(struct sock *sk) 110 { 111 u16 new_hash = ipv6_portaddr_hash(sock_net(sk), 112 &sk->sk_v6_rcv_saddr, 113 inet_sk(sk)->inet_num); 114 115 udp_lib_rehash(sk, new_hash); 116 } 117 118 static int compute_score(struct sock *sk, struct net *net, 119 const struct in6_addr *saddr, __be16 sport, 120 const struct in6_addr *daddr, unsigned short hnum, 121 int dif, int sdif) 122 { 123 int bound_dev_if, score; 124 struct inet_sock *inet; 125 bool dev_match; 126 127 if (!net_eq(sock_net(sk), net) || 128 udp_sk(sk)->udp_port_hash != hnum || 129 sk->sk_family != PF_INET6) 130 return -1; 131 132 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) 133 return -1; 134 135 score = 0; 136 inet = inet_sk(sk); 137 138 if (inet->inet_dport) { 139 if (inet->inet_dport != sport) 140 return -1; 141 score++; 142 } 143 144 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 145 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) 146 return -1; 147 score++; 148 } 149 150 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 151 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif); 152 if (!dev_match) 153 return -1; 154 if (bound_dev_if) 155 score++; 156 157 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id()) 158 score++; 159 160 return score; 161 } 162 163 /* called with rcu_read_lock() */ 164 static struct sock *udp6_lib_lookup2(struct net *net, 165 const struct in6_addr *saddr, __be16 sport, 166 const struct in6_addr *daddr, unsigned int hnum, 167 int dif, int sdif, struct udp_hslot *hslot2, 168 struct sk_buff *skb) 169 { 170 struct sock *sk, *result; 171 int score, badness; 172 bool need_rescore; 173 174 result = NULL; 175 badness = -1; 176 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 177 need_rescore = false; 178 rescore: 179 score = compute_score(need_rescore ? result : sk, net, saddr, 180 sport, daddr, hnum, dif, sdif); 181 if (score > badness) { 182 badness = score; 183 184 if (need_rescore) 185 continue; 186 187 if (sk->sk_state == TCP_ESTABLISHED) { 188 result = sk; 189 continue; 190 } 191 192 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr), 193 saddr, sport, daddr, hnum, udp6_ehashfn); 194 if (!result) { 195 result = sk; 196 continue; 197 } 198 199 /* Fall back to scoring if group has connections */ 200 if (!reuseport_has_conns(sk)) 201 return result; 202 203 /* Reuseport logic returned an error, keep original score. */ 204 if (IS_ERR(result)) 205 continue; 206 207 /* compute_score is too long of a function to be 208 * inlined, and calling it again here yields 209 * measureable overhead for some 210 * workloads. Work around it by jumping 211 * backwards to rescore 'result'. 212 */ 213 need_rescore = true; 214 goto rescore; 215 } 216 } 217 return result; 218 } 219 220 /* rcu_read_lock() must be held */ 221 struct sock *__udp6_lib_lookup(struct net *net, 222 const struct in6_addr *saddr, __be16 sport, 223 const struct in6_addr *daddr, __be16 dport, 224 int dif, int sdif, struct udp_table *udptable, 225 struct sk_buff *skb) 226 { 227 unsigned short hnum = ntohs(dport); 228 unsigned int hash2, slot2; 229 struct udp_hslot *hslot2; 230 struct sock *result, *sk; 231 232 hash2 = ipv6_portaddr_hash(net, daddr, hnum); 233 slot2 = hash2 & udptable->mask; 234 hslot2 = &udptable->hash2[slot2]; 235 236 /* Lookup connected or non-wildcard sockets */ 237 result = udp6_lib_lookup2(net, saddr, sport, 238 daddr, hnum, dif, sdif, 239 hslot2, skb); 240 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED) 241 goto done; 242 243 /* Lookup redirect from BPF */ 244 if (static_branch_unlikely(&bpf_sk_lookup_enabled) && 245 udptable == net->ipv4.udp_table) { 246 sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr), 247 saddr, sport, daddr, hnum, dif, 248 udp6_ehashfn); 249 if (sk) { 250 result = sk; 251 goto done; 252 } 253 } 254 255 /* Got non-wildcard socket or error on first lookup */ 256 if (result) 257 goto done; 258 259 /* Lookup wildcard sockets */ 260 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum); 261 slot2 = hash2 & udptable->mask; 262 hslot2 = &udptable->hash2[slot2]; 263 264 result = udp6_lib_lookup2(net, saddr, sport, 265 &in6addr_any, hnum, dif, sdif, 266 hslot2, skb); 267 done: 268 if (IS_ERR(result)) 269 return NULL; 270 return result; 271 } 272 EXPORT_SYMBOL_GPL(__udp6_lib_lookup); 273 274 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, 275 __be16 sport, __be16 dport, 276 struct udp_table *udptable) 277 { 278 const struct ipv6hdr *iph = ipv6_hdr(skb); 279 280 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, 281 &iph->daddr, dport, inet6_iif(skb), 282 inet6_sdif(skb), udptable, skb); 283 } 284 285 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb, 286 __be16 sport, __be16 dport) 287 { 288 const struct ipv6hdr *iph = ipv6_hdr(skb); 289 struct net *net = dev_net(skb->dev); 290 int iif, sdif; 291 292 inet6_get_iif_sdif(skb, &iif, &sdif); 293 294 return __udp6_lib_lookup(net, &iph->saddr, sport, 295 &iph->daddr, dport, iif, 296 sdif, net->ipv4.udp_table, NULL); 297 } 298 299 /* Must be called under rcu_read_lock(). 300 * Does increment socket refcount. 301 */ 302 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6) 303 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, 304 const struct in6_addr *daddr, __be16 dport, int dif) 305 { 306 struct sock *sk; 307 308 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, 309 dif, 0, net->ipv4.udp_table, NULL); 310 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) 311 sk = NULL; 312 return sk; 313 } 314 EXPORT_SYMBOL_GPL(udp6_lib_lookup); 315 #endif 316 317 /* do not use the scratch area len for jumbogram: their length execeeds the 318 * scratch area space; note that the IP6CB flags is still in the first 319 * cacheline, so checking for jumbograms is cheap 320 */ 321 static int udp6_skb_len(struct sk_buff *skb) 322 { 323 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); 324 } 325 326 /* 327 * This should be easy, if there is something there we 328 * return it, otherwise we block. 329 */ 330 331 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, 332 int flags, int *addr_len) 333 { 334 struct ipv6_pinfo *np = inet6_sk(sk); 335 struct inet_sock *inet = inet_sk(sk); 336 struct sk_buff *skb; 337 unsigned int ulen, copied; 338 int off, err, peeking = flags & MSG_PEEK; 339 int is_udplite = IS_UDPLITE(sk); 340 struct udp_mib __percpu *mib; 341 bool checksum_valid = false; 342 int is_udp4; 343 344 if (flags & MSG_ERRQUEUE) 345 return ipv6_recv_error(sk, msg, len, addr_len); 346 347 if (np->rxpmtu && np->rxopt.bits.rxpmtu) 348 return ipv6_recv_rxpmtu(sk, msg, len, addr_len); 349 350 try_again: 351 off = sk_peek_offset(sk, flags); 352 skb = __skb_recv_udp(sk, flags, &off, &err); 353 if (!skb) 354 return err; 355 356 ulen = udp6_skb_len(skb); 357 copied = len; 358 if (copied > ulen - off) 359 copied = ulen - off; 360 else if (copied < ulen) 361 msg->msg_flags |= MSG_TRUNC; 362 363 is_udp4 = (skb->protocol == htons(ETH_P_IP)); 364 mib = __UDPX_MIB(sk, is_udp4); 365 366 /* 367 * If checksum is needed at all, try to do it while copying the 368 * data. If the data is truncated, or if we only want a partial 369 * coverage checksum (UDP-Lite), do it before the copy. 370 */ 371 372 if (copied < ulen || peeking || 373 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { 374 checksum_valid = udp_skb_csum_unnecessary(skb) || 375 !__udp_lib_checksum_complete(skb); 376 if (!checksum_valid) 377 goto csum_copy_err; 378 } 379 380 if (checksum_valid || udp_skb_csum_unnecessary(skb)) { 381 if (udp_skb_is_linear(skb)) 382 err = copy_linear_skb(skb, copied, off, &msg->msg_iter); 383 else 384 err = skb_copy_datagram_msg(skb, off, msg, copied); 385 } else { 386 err = skb_copy_and_csum_datagram_msg(skb, off, msg); 387 if (err == -EINVAL) 388 goto csum_copy_err; 389 } 390 if (unlikely(err)) { 391 if (!peeking) { 392 atomic_inc(&sk->sk_drops); 393 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 394 } 395 kfree_skb(skb); 396 return err; 397 } 398 if (!peeking) 399 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS); 400 401 sock_recv_cmsgs(msg, sk, skb); 402 403 /* Copy the address. */ 404 if (msg->msg_name) { 405 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 406 sin6->sin6_family = AF_INET6; 407 sin6->sin6_port = udp_hdr(skb)->source; 408 sin6->sin6_flowinfo = 0; 409 410 if (is_udp4) { 411 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, 412 &sin6->sin6_addr); 413 sin6->sin6_scope_id = 0; 414 } else { 415 sin6->sin6_addr = ipv6_hdr(skb)->saddr; 416 sin6->sin6_scope_id = 417 ipv6_iface_scope_id(&sin6->sin6_addr, 418 inet6_iif(skb)); 419 } 420 *addr_len = sizeof(*sin6); 421 422 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, 423 (struct sockaddr *)sin6, 424 addr_len); 425 } 426 427 if (udp_test_bit(GRO_ENABLED, sk)) 428 udp_cmsg_recv(msg, sk, skb); 429 430 if (np->rxopt.all) 431 ip6_datagram_recv_common_ctl(sk, msg, skb); 432 433 if (is_udp4) { 434 if (inet_cmsg_flags(inet)) 435 ip_cmsg_recv_offset(msg, sk, skb, 436 sizeof(struct udphdr), off); 437 } else { 438 if (np->rxopt.all) 439 ip6_datagram_recv_specific_ctl(sk, msg, skb); 440 } 441 442 err = copied; 443 if (flags & MSG_TRUNC) 444 err = ulen; 445 446 skb_consume_udp(sk, skb, peeking ? -err : err); 447 return err; 448 449 csum_copy_err: 450 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, 451 udp_skb_destructor)) { 452 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS); 453 SNMP_INC_STATS(mib, UDP_MIB_INERRORS); 454 } 455 kfree_skb(skb); 456 457 /* starting over for a new packet, but check if we need to yield */ 458 cond_resched(); 459 msg->msg_flags &= ~MSG_TRUNC; 460 goto try_again; 461 } 462 463 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key); 464 void udpv6_encap_enable(void) 465 { 466 static_branch_inc(&udpv6_encap_needed_key); 467 } 468 EXPORT_SYMBOL(udpv6_encap_enable); 469 470 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go 471 * through error handlers in encapsulations looking for a match. 472 */ 473 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb, 474 struct inet6_skb_parm *opt, 475 u8 type, u8 code, int offset, __be32 info) 476 { 477 int i; 478 479 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) { 480 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt, 481 u8 type, u8 code, int offset, __be32 info); 482 const struct ip6_tnl_encap_ops *encap; 483 484 encap = rcu_dereference(ip6tun_encaps[i]); 485 if (!encap) 486 continue; 487 handler = encap->err_handler; 488 if (handler && !handler(skb, opt, type, code, offset, info)) 489 return 0; 490 } 491 492 return -ENOENT; 493 } 494 495 /* Try to match ICMP errors to UDP tunnels by looking up a socket without 496 * reversing source and destination port: this will match tunnels that force the 497 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that 498 * lwtunnels might actually break this assumption by being configured with 499 * different destination ports on endpoints, in this case we won't be able to 500 * trace ICMP messages back to them. 501 * 502 * If this doesn't match any socket, probe tunnels with arbitrary destination 503 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port 504 * we've sent packets to won't necessarily match the local destination port. 505 * 506 * Then ask the tunnel implementation to match the error against a valid 507 * association. 508 * 509 * Return an error if we can't find a match, the socket if we need further 510 * processing, zero otherwise. 511 */ 512 static struct sock *__udp6_lib_err_encap(struct net *net, 513 const struct ipv6hdr *hdr, int offset, 514 struct udphdr *uh, 515 struct udp_table *udptable, 516 struct sock *sk, 517 struct sk_buff *skb, 518 struct inet6_skb_parm *opt, 519 u8 type, u8 code, __be32 info) 520 { 521 int (*lookup)(struct sock *sk, struct sk_buff *skb); 522 int network_offset, transport_offset; 523 struct udp_sock *up; 524 525 network_offset = skb_network_offset(skb); 526 transport_offset = skb_transport_offset(skb); 527 528 /* Network header needs to point to the outer IPv6 header inside ICMP */ 529 skb_reset_network_header(skb); 530 531 /* Transport header needs to point to the UDP header */ 532 skb_set_transport_header(skb, offset); 533 534 if (sk) { 535 up = udp_sk(sk); 536 537 lookup = READ_ONCE(up->encap_err_lookup); 538 if (lookup && lookup(sk, skb)) 539 sk = NULL; 540 541 goto out; 542 } 543 544 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source, 545 &hdr->saddr, uh->dest, 546 inet6_iif(skb), 0, udptable, skb); 547 if (sk) { 548 up = udp_sk(sk); 549 550 lookup = READ_ONCE(up->encap_err_lookup); 551 if (!lookup || lookup(sk, skb)) 552 sk = NULL; 553 } 554 555 out: 556 if (!sk) { 557 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code, 558 offset, info)); 559 } 560 561 skb_set_transport_header(skb, transport_offset); 562 skb_set_network_header(skb, network_offset); 563 564 return sk; 565 } 566 567 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, 568 u8 type, u8 code, int offset, __be32 info, 569 struct udp_table *udptable) 570 { 571 struct ipv6_pinfo *np; 572 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; 573 const struct in6_addr *saddr = &hdr->saddr; 574 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr; 575 struct udphdr *uh = (struct udphdr *)(skb->data+offset); 576 bool tunnel = false; 577 struct sock *sk; 578 int harderr; 579 int err; 580 struct net *net = dev_net(skb->dev); 581 582 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, 583 inet6_iif(skb), inet6_sdif(skb), udptable, NULL); 584 585 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) { 586 /* No socket for error: try tunnels before discarding */ 587 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 588 sk = __udp6_lib_err_encap(net, hdr, offset, uh, 589 udptable, sk, skb, 590 opt, type, code, info); 591 if (!sk) 592 return 0; 593 } else 594 sk = ERR_PTR(-ENOENT); 595 596 if (IS_ERR(sk)) { 597 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), 598 ICMP6_MIB_INERRORS); 599 return PTR_ERR(sk); 600 } 601 602 tunnel = true; 603 } 604 605 harderr = icmpv6_err_convert(type, code, &err); 606 np = inet6_sk(sk); 607 608 if (type == ICMPV6_PKT_TOOBIG) { 609 if (!ip6_sk_accept_pmtu(sk)) 610 goto out; 611 ip6_sk_update_pmtu(skb, sk, info); 612 if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT) 613 harderr = 1; 614 } 615 if (type == NDISC_REDIRECT) { 616 if (tunnel) { 617 ip6_redirect(skb, sock_net(sk), inet6_iif(skb), 618 READ_ONCE(sk->sk_mark), sk->sk_uid); 619 } else { 620 ip6_sk_redirect(skb, sk); 621 } 622 goto out; 623 } 624 625 /* Tunnels don't have an application socket: don't pass errors back */ 626 if (tunnel) { 627 if (udp_sk(sk)->encap_err_rcv) 628 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, 629 ntohl(info), (u8 *)(uh+1)); 630 goto out; 631 } 632 633 if (!inet6_test_bit(RECVERR6, sk)) { 634 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 635 goto out; 636 } else { 637 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); 638 } 639 640 sk->sk_err = err; 641 sk_error_report(sk); 642 out: 643 return 0; 644 } 645 646 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 647 { 648 int rc; 649 650 if (!ipv6_addr_any(&sk->sk_v6_daddr)) { 651 sock_rps_save_rxhash(sk, skb); 652 sk_mark_napi_id(sk, skb); 653 sk_incoming_cpu_update(sk); 654 } else { 655 sk_mark_napi_id_once(sk, skb); 656 } 657 658 rc = __udp_enqueue_schedule_skb(sk, skb); 659 if (rc < 0) { 660 int is_udplite = IS_UDPLITE(sk); 661 enum skb_drop_reason drop_reason; 662 663 /* Note that an ENOMEM error is charged twice */ 664 if (rc == -ENOMEM) { 665 UDP6_INC_STATS(sock_net(sk), 666 UDP_MIB_RCVBUFERRORS, is_udplite); 667 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 668 } else { 669 UDP6_INC_STATS(sock_net(sk), 670 UDP_MIB_MEMERRORS, is_udplite); 671 drop_reason = SKB_DROP_REASON_PROTO_MEM; 672 } 673 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 674 trace_udp_fail_queue_rcv_skb(rc, sk, skb); 675 kfree_skb_reason(skb, drop_reason); 676 return -1; 677 } 678 679 return 0; 680 } 681 682 static __inline__ int udpv6_err(struct sk_buff *skb, 683 struct inet6_skb_parm *opt, u8 type, 684 u8 code, int offset, __be32 info) 685 { 686 return __udp6_lib_err(skb, opt, type, code, offset, info, 687 dev_net(skb->dev)->ipv4.udp_table); 688 } 689 690 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb) 691 { 692 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; 693 struct udp_sock *up = udp_sk(sk); 694 int is_udplite = IS_UDPLITE(sk); 695 696 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) { 697 drop_reason = SKB_DROP_REASON_XFRM_POLICY; 698 goto drop; 699 } 700 nf_reset_ct(skb); 701 702 if (static_branch_unlikely(&udpv6_encap_needed_key) && 703 READ_ONCE(up->encap_type)) { 704 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); 705 706 /* 707 * This is an encapsulation socket so pass the skb to 708 * the socket's udp_encap_rcv() hook. Otherwise, just 709 * fall through and pass this up the UDP socket. 710 * up->encap_rcv() returns the following value: 711 * =0 if skb was successfully passed to the encap 712 * handler or was discarded by it. 713 * >0 if skb should be passed on to UDP. 714 * <0 if skb should be resubmitted as proto -N 715 */ 716 717 /* if we're overly short, let UDP handle it */ 718 encap_rcv = READ_ONCE(up->encap_rcv); 719 if (encap_rcv) { 720 int ret; 721 722 /* Verify checksum before giving to encap */ 723 if (udp_lib_checksum_complete(skb)) 724 goto csum_error; 725 726 ret = encap_rcv(sk, skb); 727 if (ret <= 0) { 728 __UDP6_INC_STATS(sock_net(sk), 729 UDP_MIB_INDATAGRAMS, 730 is_udplite); 731 return -ret; 732 } 733 } 734 735 /* FALLTHROUGH -- it's a UDP Packet */ 736 } 737 738 /* 739 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). 740 */ 741 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) { 742 u16 pcrlen = READ_ONCE(up->pcrlen); 743 744 if (pcrlen == 0) { /* full coverage was set */ 745 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", 746 UDP_SKB_CB(skb)->cscov, skb->len); 747 goto drop; 748 } 749 if (UDP_SKB_CB(skb)->cscov < pcrlen) { 750 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", 751 UDP_SKB_CB(skb)->cscov, pcrlen); 752 goto drop; 753 } 754 } 755 756 prefetch(&sk->sk_rmem_alloc); 757 if (rcu_access_pointer(sk->sk_filter) && 758 udp_lib_checksum_complete(skb)) 759 goto csum_error; 760 761 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) { 762 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 763 goto drop; 764 } 765 766 udp_csum_pull_header(skb); 767 768 skb_dst_drop(skb); 769 770 return __udpv6_queue_rcv_skb(sk, skb); 771 772 csum_error: 773 drop_reason = SKB_DROP_REASON_UDP_CSUM; 774 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); 775 drop: 776 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 777 atomic_inc(&sk->sk_drops); 778 kfree_skb_reason(skb, drop_reason); 779 return -1; 780 } 781 782 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 783 { 784 struct sk_buff *next, *segs; 785 int ret; 786 787 if (likely(!udp_unexpected_gso(sk, skb))) 788 return udpv6_queue_rcv_one_skb(sk, skb); 789 790 __skb_push(skb, -skb_mac_offset(skb)); 791 segs = udp_rcv_segment(sk, skb, false); 792 skb_list_walk_safe(segs, skb, next) { 793 __skb_pull(skb, skb_transport_offset(skb)); 794 795 udp_post_segment_fix_csum(skb); 796 ret = udpv6_queue_rcv_one_skb(sk, skb); 797 if (ret > 0) 798 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret, 799 true); 800 } 801 return 0; 802 } 803 804 static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk, 805 __be16 loc_port, const struct in6_addr *loc_addr, 806 __be16 rmt_port, const struct in6_addr *rmt_addr, 807 int dif, int sdif, unsigned short hnum) 808 { 809 const struct inet_sock *inet = inet_sk(sk); 810 811 if (!net_eq(sock_net(sk), net)) 812 return false; 813 814 if (udp_sk(sk)->udp_port_hash != hnum || 815 sk->sk_family != PF_INET6 || 816 (inet->inet_dport && inet->inet_dport != rmt_port) || 817 (!ipv6_addr_any(&sk->sk_v6_daddr) && 818 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || 819 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) || 820 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && 821 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) 822 return false; 823 if (!inet6_mc_check(sk, loc_addr, rmt_addr)) 824 return false; 825 return true; 826 } 827 828 static void udp6_csum_zero_error(struct sk_buff *skb) 829 { 830 /* RFC 2460 section 8.1 says that we SHOULD log 831 * this error. Well, it is reasonable. 832 */ 833 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", 834 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), 835 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); 836 } 837 838 /* 839 * Note: called only from the BH handler context, 840 * so we don't need to lock the hashes. 841 */ 842 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 843 const struct in6_addr *saddr, const struct in6_addr *daddr, 844 struct udp_table *udptable, int proto) 845 { 846 struct sock *sk, *first = NULL; 847 const struct udphdr *uh = udp_hdr(skb); 848 unsigned short hnum = ntohs(uh->dest); 849 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); 850 unsigned int offset = offsetof(typeof(*sk), sk_node); 851 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); 852 int dif = inet6_iif(skb); 853 int sdif = inet6_sdif(skb); 854 struct hlist_node *node; 855 struct sk_buff *nskb; 856 857 if (use_hash2) { 858 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) & 859 udptable->mask; 860 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask; 861 start_lookup: 862 hslot = &udptable->hash2[hash2]; 863 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); 864 } 865 866 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { 867 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, 868 uh->source, saddr, dif, sdif, 869 hnum)) 870 continue; 871 /* If zero checksum and no_check is not on for 872 * the socket then skip it. 873 */ 874 if (!uh->check && !udp_get_no_check6_rx(sk)) 875 continue; 876 if (!first) { 877 first = sk; 878 continue; 879 } 880 nskb = skb_clone(skb, GFP_ATOMIC); 881 if (unlikely(!nskb)) { 882 atomic_inc(&sk->sk_drops); 883 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, 884 IS_UDPLITE(sk)); 885 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, 886 IS_UDPLITE(sk)); 887 continue; 888 } 889 890 if (udpv6_queue_rcv_skb(sk, nskb) > 0) 891 consume_skb(nskb); 892 } 893 894 /* Also lookup *:port if we are using hash2 and haven't done so yet. */ 895 if (use_hash2 && hash2 != hash2_any) { 896 hash2 = hash2_any; 897 goto start_lookup; 898 } 899 900 if (first) { 901 if (udpv6_queue_rcv_skb(first, skb) > 0) 902 consume_skb(skb); 903 } else { 904 kfree_skb(skb); 905 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, 906 proto == IPPROTO_UDPLITE); 907 } 908 return 0; 909 } 910 911 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) 912 { 913 if (udp_sk_rx_dst_set(sk, dst)) { 914 const struct rt6_info *rt = (const struct rt6_info *)dst; 915 916 sk->sk_rx_dst_cookie = rt6_get_cookie(rt); 917 } 918 } 919 920 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and 921 * return code conversion for ip layer consumption 922 */ 923 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb, 924 struct udphdr *uh) 925 { 926 int ret; 927 928 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) 929 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo); 930 931 ret = udpv6_queue_rcv_skb(sk, skb); 932 933 /* a return value > 0 means to resubmit the input */ 934 if (ret > 0) 935 return ret; 936 return 0; 937 } 938 939 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, 940 int proto) 941 { 942 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED; 943 const struct in6_addr *saddr, *daddr; 944 struct net *net = dev_net(skb->dev); 945 struct udphdr *uh; 946 struct sock *sk; 947 bool refcounted; 948 u32 ulen = 0; 949 950 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 951 goto discard; 952 953 saddr = &ipv6_hdr(skb)->saddr; 954 daddr = &ipv6_hdr(skb)->daddr; 955 uh = udp_hdr(skb); 956 957 ulen = ntohs(uh->len); 958 if (ulen > skb->len) 959 goto short_packet; 960 961 if (proto == IPPROTO_UDP) { 962 /* UDP validates ulen. */ 963 964 /* Check for jumbo payload */ 965 if (ulen == 0) 966 ulen = skb->len; 967 968 if (ulen < sizeof(*uh)) 969 goto short_packet; 970 971 if (ulen < skb->len) { 972 if (pskb_trim_rcsum(skb, ulen)) 973 goto short_packet; 974 saddr = &ipv6_hdr(skb)->saddr; 975 daddr = &ipv6_hdr(skb)->daddr; 976 uh = udp_hdr(skb); 977 } 978 } 979 980 if (udp6_csum_init(skb, uh, proto)) 981 goto csum_error; 982 983 /* Check if the socket is already available, e.g. due to early demux */ 984 sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest, 985 &refcounted, udp6_ehashfn); 986 if (IS_ERR(sk)) 987 goto no_sk; 988 989 if (sk) { 990 struct dst_entry *dst = skb_dst(skb); 991 int ret; 992 993 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst)) 994 udp6_sk_rx_dst_set(sk, dst); 995 996 if (!uh->check && !udp_get_no_check6_rx(sk)) { 997 if (refcounted) 998 sock_put(sk); 999 goto report_csum_error; 1000 } 1001 1002 ret = udp6_unicast_rcv_skb(sk, skb, uh); 1003 if (refcounted) 1004 sock_put(sk); 1005 return ret; 1006 } 1007 1008 /* 1009 * Multicast receive code 1010 */ 1011 if (ipv6_addr_is_multicast(daddr)) 1012 return __udp6_lib_mcast_deliver(net, skb, 1013 saddr, daddr, udptable, proto); 1014 1015 /* Unicast */ 1016 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1017 if (sk) { 1018 if (!uh->check && !udp_get_no_check6_rx(sk)) 1019 goto report_csum_error; 1020 return udp6_unicast_rcv_skb(sk, skb, uh); 1021 } 1022 no_sk: 1023 reason = SKB_DROP_REASON_NO_SOCKET; 1024 1025 if (!uh->check) 1026 goto report_csum_error; 1027 1028 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) 1029 goto discard; 1030 nf_reset_ct(skb); 1031 1032 if (udp_lib_checksum_complete(skb)) 1033 goto csum_error; 1034 1035 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1036 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); 1037 1038 kfree_skb_reason(skb, reason); 1039 return 0; 1040 1041 short_packet: 1042 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1043 reason = SKB_DROP_REASON_PKT_TOO_SMALL; 1044 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", 1045 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1046 saddr, ntohs(uh->source), 1047 ulen, skb->len, 1048 daddr, ntohs(uh->dest)); 1049 goto discard; 1050 1051 report_csum_error: 1052 udp6_csum_zero_error(skb); 1053 csum_error: 1054 if (reason == SKB_DROP_REASON_NOT_SPECIFIED) 1055 reason = SKB_DROP_REASON_UDP_CSUM; 1056 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); 1057 discard: 1058 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1059 kfree_skb_reason(skb, reason); 1060 return 0; 1061 } 1062 1063 1064 static struct sock *__udp6_lib_demux_lookup(struct net *net, 1065 __be16 loc_port, const struct in6_addr *loc_addr, 1066 __be16 rmt_port, const struct in6_addr *rmt_addr, 1067 int dif, int sdif) 1068 { 1069 struct udp_table *udptable = net->ipv4.udp_table; 1070 unsigned short hnum = ntohs(loc_port); 1071 unsigned int hash2, slot2; 1072 struct udp_hslot *hslot2; 1073 __portpair ports; 1074 struct sock *sk; 1075 1076 hash2 = ipv6_portaddr_hash(net, loc_addr, hnum); 1077 slot2 = hash2 & udptable->mask; 1078 hslot2 = &udptable->hash2[slot2]; 1079 ports = INET_COMBINED_PORTS(rmt_port, hnum); 1080 1081 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { 1082 if (sk->sk_state == TCP_ESTABLISHED && 1083 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif)) 1084 return sk; 1085 /* Only check first socket in chain */ 1086 break; 1087 } 1088 return NULL; 1089 } 1090 1091 void udp_v6_early_demux(struct sk_buff *skb) 1092 { 1093 struct net *net = dev_net(skb->dev); 1094 const struct udphdr *uh; 1095 struct sock *sk; 1096 struct dst_entry *dst; 1097 int dif = skb->dev->ifindex; 1098 int sdif = inet6_sdif(skb); 1099 1100 if (!pskb_may_pull(skb, skb_transport_offset(skb) + 1101 sizeof(struct udphdr))) 1102 return; 1103 1104 uh = udp_hdr(skb); 1105 1106 if (skb->pkt_type == PACKET_HOST) 1107 sk = __udp6_lib_demux_lookup(net, uh->dest, 1108 &ipv6_hdr(skb)->daddr, 1109 uh->source, &ipv6_hdr(skb)->saddr, 1110 dif, sdif); 1111 else 1112 return; 1113 1114 if (!sk) 1115 return; 1116 1117 skb->sk = sk; 1118 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk)); 1119 skb->destructor = sock_pfree; 1120 dst = rcu_dereference(sk->sk_rx_dst); 1121 1122 if (dst) 1123 dst = dst_check(dst, sk->sk_rx_dst_cookie); 1124 if (dst) { 1125 /* set noref for now. 1126 * any place which wants to hold dst has to call 1127 * dst_hold_safe() 1128 */ 1129 skb_dst_set_noref(skb, dst); 1130 } 1131 } 1132 1133 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb) 1134 { 1135 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP); 1136 } 1137 1138 /* 1139 * Throw away all pending data and cancel the corking. Socket is locked. 1140 */ 1141 static void udp_v6_flush_pending_frames(struct sock *sk) 1142 { 1143 struct udp_sock *up = udp_sk(sk); 1144 1145 if (up->pending == AF_INET) 1146 udp_flush_pending_frames(sk); 1147 else if (up->pending) { 1148 up->len = 0; 1149 WRITE_ONCE(up->pending, 0); 1150 ip6_flush_pending_frames(sk); 1151 } 1152 } 1153 1154 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr, 1155 int addr_len) 1156 { 1157 if (addr_len < offsetofend(struct sockaddr, sa_family)) 1158 return -EINVAL; 1159 /* The following checks are replicated from __ip6_datagram_connect() 1160 * and intended to prevent BPF program called below from accessing 1161 * bytes that are out of the bound specified by user in addr_len. 1162 */ 1163 if (uaddr->sa_family == AF_INET) { 1164 if (ipv6_only_sock(sk)) 1165 return -EAFNOSUPPORT; 1166 return udp_pre_connect(sk, uaddr, addr_len); 1167 } 1168 1169 if (addr_len < SIN6_LEN_RFC2133) 1170 return -EINVAL; 1171 1172 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len); 1173 } 1174 1175 /** 1176 * udp6_hwcsum_outgoing - handle outgoing HW checksumming 1177 * @sk: socket we are sending on 1178 * @skb: sk_buff containing the filled-in UDP header 1179 * (checksum field must be zeroed out) 1180 * @saddr: source address 1181 * @daddr: destination address 1182 * @len: length of packet 1183 */ 1184 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 1185 const struct in6_addr *saddr, 1186 const struct in6_addr *daddr, int len) 1187 { 1188 unsigned int offset; 1189 struct udphdr *uh = udp_hdr(skb); 1190 struct sk_buff *frags = skb_shinfo(skb)->frag_list; 1191 __wsum csum = 0; 1192 1193 if (!frags) { 1194 /* Only one fragment on the socket. */ 1195 skb->csum_start = skb_transport_header(skb) - skb->head; 1196 skb->csum_offset = offsetof(struct udphdr, check); 1197 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); 1198 } else { 1199 /* 1200 * HW-checksum won't work as there are two or more 1201 * fragments on the socket so that all csums of sk_buffs 1202 * should be together 1203 */ 1204 offset = skb_transport_offset(skb); 1205 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 1206 csum = skb->csum; 1207 1208 skb->ip_summed = CHECKSUM_NONE; 1209 1210 do { 1211 csum = csum_add(csum, frags->csum); 1212 } while ((frags = frags->next)); 1213 1214 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 1215 csum); 1216 if (uh->check == 0) 1217 uh->check = CSUM_MANGLED_0; 1218 } 1219 } 1220 1221 /* 1222 * Sending 1223 */ 1224 1225 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6, 1226 struct inet_cork *cork) 1227 { 1228 struct sock *sk = skb->sk; 1229 struct udphdr *uh; 1230 int err = 0; 1231 int is_udplite = IS_UDPLITE(sk); 1232 __wsum csum = 0; 1233 int offset = skb_transport_offset(skb); 1234 int len = skb->len - offset; 1235 int datalen = len - sizeof(*uh); 1236 1237 /* 1238 * Create a UDP header 1239 */ 1240 uh = udp_hdr(skb); 1241 uh->source = fl6->fl6_sport; 1242 uh->dest = fl6->fl6_dport; 1243 uh->len = htons(len); 1244 uh->check = 0; 1245 1246 if (cork->gso_size) { 1247 const int hlen = skb_network_header_len(skb) + 1248 sizeof(struct udphdr); 1249 1250 if (hlen + cork->gso_size > cork->fragsize) { 1251 kfree_skb(skb); 1252 return -EINVAL; 1253 } 1254 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) { 1255 kfree_skb(skb); 1256 return -EINVAL; 1257 } 1258 if (udp_get_no_check6_tx(sk)) { 1259 kfree_skb(skb); 1260 return -EINVAL; 1261 } 1262 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite || 1263 dst_xfrm(skb_dst(skb))) { 1264 kfree_skb(skb); 1265 return -EIO; 1266 } 1267 1268 if (datalen > cork->gso_size) { 1269 skb_shinfo(skb)->gso_size = cork->gso_size; 1270 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4; 1271 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen, 1272 cork->gso_size); 1273 } 1274 goto csum_partial; 1275 } 1276 1277 if (is_udplite) 1278 csum = udplite_csum(skb); 1279 else if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */ 1280 skb->ip_summed = CHECKSUM_NONE; 1281 goto send; 1282 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 1283 csum_partial: 1284 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); 1285 goto send; 1286 } else 1287 csum = udp_csum(skb); 1288 1289 /* add protocol-dependent pseudo-header */ 1290 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, 1291 len, fl6->flowi6_proto, csum); 1292 if (uh->check == 0) 1293 uh->check = CSUM_MANGLED_0; 1294 1295 send: 1296 err = ip6_send_skb(skb); 1297 if (err) { 1298 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) { 1299 UDP6_INC_STATS(sock_net(sk), 1300 UDP_MIB_SNDBUFERRORS, is_udplite); 1301 err = 0; 1302 } 1303 } else { 1304 UDP6_INC_STATS(sock_net(sk), 1305 UDP_MIB_OUTDATAGRAMS, is_udplite); 1306 } 1307 return err; 1308 } 1309 1310 static int udp_v6_push_pending_frames(struct sock *sk) 1311 { 1312 struct sk_buff *skb; 1313 struct udp_sock *up = udp_sk(sk); 1314 int err = 0; 1315 1316 if (up->pending == AF_INET) 1317 return udp_push_pending_frames(sk); 1318 1319 skb = ip6_finish_skb(sk); 1320 if (!skb) 1321 goto out; 1322 1323 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6, 1324 &inet_sk(sk)->cork.base); 1325 out: 1326 up->len = 0; 1327 WRITE_ONCE(up->pending, 0); 1328 return err; 1329 } 1330 1331 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) 1332 { 1333 struct ipv6_txoptions opt_space; 1334 struct udp_sock *up = udp_sk(sk); 1335 struct inet_sock *inet = inet_sk(sk); 1336 struct ipv6_pinfo *np = inet6_sk(sk); 1337 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); 1338 struct in6_addr *daddr, *final_p, final; 1339 struct ipv6_txoptions *opt = NULL; 1340 struct ipv6_txoptions *opt_to_free = NULL; 1341 struct ip6_flowlabel *flowlabel = NULL; 1342 struct inet_cork_full cork; 1343 struct flowi6 *fl6 = &cork.fl.u.ip6; 1344 struct dst_entry *dst; 1345 struct ipcm6_cookie ipc6; 1346 int addr_len = msg->msg_namelen; 1347 bool connected = false; 1348 int ulen = len; 1349 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE; 1350 int err; 1351 int is_udplite = IS_UDPLITE(sk); 1352 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 1353 1354 ipcm6_init(&ipc6); 1355 ipc6.gso_size = READ_ONCE(up->gso_size); 1356 ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags); 1357 ipc6.sockc.mark = READ_ONCE(sk->sk_mark); 1358 1359 /* destination address check */ 1360 if (sin6) { 1361 if (addr_len < offsetof(struct sockaddr, sa_data)) 1362 return -EINVAL; 1363 1364 switch (sin6->sin6_family) { 1365 case AF_INET6: 1366 if (addr_len < SIN6_LEN_RFC2133) 1367 return -EINVAL; 1368 daddr = &sin6->sin6_addr; 1369 if (ipv6_addr_any(daddr) && 1370 ipv6_addr_v4mapped(&np->saddr)) 1371 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), 1372 daddr); 1373 break; 1374 case AF_INET: 1375 goto do_udp_sendmsg; 1376 case AF_UNSPEC: 1377 msg->msg_name = sin6 = NULL; 1378 msg->msg_namelen = addr_len = 0; 1379 daddr = NULL; 1380 break; 1381 default: 1382 return -EINVAL; 1383 } 1384 } else if (!READ_ONCE(up->pending)) { 1385 if (sk->sk_state != TCP_ESTABLISHED) 1386 return -EDESTADDRREQ; 1387 daddr = &sk->sk_v6_daddr; 1388 } else 1389 daddr = NULL; 1390 1391 if (daddr) { 1392 if (ipv6_addr_v4mapped(daddr)) { 1393 struct sockaddr_in sin; 1394 sin.sin_family = AF_INET; 1395 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; 1396 sin.sin_addr.s_addr = daddr->s6_addr32[3]; 1397 msg->msg_name = &sin; 1398 msg->msg_namelen = sizeof(sin); 1399 do_udp_sendmsg: 1400 err = ipv6_only_sock(sk) ? 1401 -ENETUNREACH : udp_sendmsg(sk, msg, len); 1402 msg->msg_name = sin6; 1403 msg->msg_namelen = addr_len; 1404 return err; 1405 } 1406 } 1407 1408 /* Rough check on arithmetic overflow, 1409 better check is made in ip6_append_data(). 1410 */ 1411 if (len > INT_MAX - sizeof(struct udphdr)) 1412 return -EMSGSIZE; 1413 1414 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 1415 if (READ_ONCE(up->pending)) { 1416 if (READ_ONCE(up->pending) == AF_INET) 1417 return udp_sendmsg(sk, msg, len); 1418 /* 1419 * There are pending frames. 1420 * The socket lock must be held while it's corked. 1421 */ 1422 lock_sock(sk); 1423 if (likely(up->pending)) { 1424 if (unlikely(up->pending != AF_INET6)) { 1425 release_sock(sk); 1426 return -EAFNOSUPPORT; 1427 } 1428 dst = NULL; 1429 goto do_append_data; 1430 } 1431 release_sock(sk); 1432 } 1433 ulen += sizeof(struct udphdr); 1434 1435 memset(fl6, 0, sizeof(*fl6)); 1436 1437 if (sin6) { 1438 if (sin6->sin6_port == 0) 1439 return -EINVAL; 1440 1441 fl6->fl6_dport = sin6->sin6_port; 1442 daddr = &sin6->sin6_addr; 1443 1444 if (inet6_test_bit(SNDFLOW, sk)) { 1445 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; 1446 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) { 1447 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1448 if (IS_ERR(flowlabel)) 1449 return -EINVAL; 1450 } 1451 } 1452 1453 /* 1454 * Otherwise it will be difficult to maintain 1455 * sk->sk_dst_cache. 1456 */ 1457 if (sk->sk_state == TCP_ESTABLISHED && 1458 ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) 1459 daddr = &sk->sk_v6_daddr; 1460 1461 if (addr_len >= sizeof(struct sockaddr_in6) && 1462 sin6->sin6_scope_id && 1463 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) 1464 fl6->flowi6_oif = sin6->sin6_scope_id; 1465 } else { 1466 if (sk->sk_state != TCP_ESTABLISHED) 1467 return -EDESTADDRREQ; 1468 1469 fl6->fl6_dport = inet->inet_dport; 1470 daddr = &sk->sk_v6_daddr; 1471 fl6->flowlabel = np->flow_label; 1472 connected = true; 1473 } 1474 1475 if (!fl6->flowi6_oif) 1476 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if); 1477 1478 if (!fl6->flowi6_oif) 1479 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; 1480 1481 fl6->flowi6_uid = sk->sk_uid; 1482 1483 if (msg->msg_controllen) { 1484 opt = &opt_space; 1485 memset(opt, 0, sizeof(struct ipv6_txoptions)); 1486 opt->tot_len = sizeof(*opt); 1487 ipc6.opt = opt; 1488 1489 err = udp_cmsg_send(sk, msg, &ipc6.gso_size); 1490 if (err > 0) 1491 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6, 1492 &ipc6); 1493 if (err < 0) { 1494 fl6_sock_release(flowlabel); 1495 return err; 1496 } 1497 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { 1498 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel); 1499 if (IS_ERR(flowlabel)) 1500 return -EINVAL; 1501 } 1502 if (!(opt->opt_nflen|opt->opt_flen)) 1503 opt = NULL; 1504 connected = false; 1505 } 1506 if (!opt) { 1507 opt = txopt_get(np); 1508 opt_to_free = opt; 1509 } 1510 if (flowlabel) 1511 opt = fl6_merge_options(&opt_space, flowlabel, opt); 1512 opt = ipv6_fixup_options(&opt_space, opt); 1513 ipc6.opt = opt; 1514 1515 fl6->flowi6_proto = sk->sk_protocol; 1516 fl6->flowi6_mark = ipc6.sockc.mark; 1517 fl6->daddr = *daddr; 1518 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr)) 1519 fl6->saddr = np->saddr; 1520 fl6->fl6_sport = inet->inet_sport; 1521 1522 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) { 1523 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, 1524 (struct sockaddr *)sin6, 1525 &addr_len, 1526 &fl6->saddr); 1527 if (err) 1528 goto out_no_dst; 1529 if (sin6) { 1530 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) { 1531 /* BPF program rewrote IPv6-only by IPv4-mapped 1532 * IPv6. It's currently unsupported. 1533 */ 1534 err = -ENOTSUPP; 1535 goto out_no_dst; 1536 } 1537 if (sin6->sin6_port == 0) { 1538 /* BPF program set invalid port. Reject it. */ 1539 err = -EINVAL; 1540 goto out_no_dst; 1541 } 1542 fl6->fl6_dport = sin6->sin6_port; 1543 fl6->daddr = sin6->sin6_addr; 1544 } 1545 } 1546 1547 if (ipv6_addr_any(&fl6->daddr)) 1548 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ 1549 1550 final_p = fl6_update_dst(fl6, opt, &final); 1551 if (final_p) 1552 connected = false; 1553 1554 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) { 1555 fl6->flowi6_oif = READ_ONCE(np->mcast_oif); 1556 connected = false; 1557 } else if (!fl6->flowi6_oif) 1558 fl6->flowi6_oif = READ_ONCE(np->ucast_oif); 1559 1560 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6)); 1561 1562 if (ipc6.tclass < 0) 1563 ipc6.tclass = np->tclass; 1564 1565 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel); 1566 1567 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected); 1568 if (IS_ERR(dst)) { 1569 err = PTR_ERR(dst); 1570 dst = NULL; 1571 goto out; 1572 } 1573 1574 if (ipc6.hlimit < 0) 1575 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst); 1576 1577 if (msg->msg_flags&MSG_CONFIRM) 1578 goto do_confirm; 1579 back_from_confirm: 1580 1581 /* Lockless fast path for the non-corking case */ 1582 if (!corkreq) { 1583 struct sk_buff *skb; 1584 1585 skb = ip6_make_skb(sk, getfrag, msg, ulen, 1586 sizeof(struct udphdr), &ipc6, 1587 (struct rt6_info *)dst, 1588 msg->msg_flags, &cork); 1589 err = PTR_ERR(skb); 1590 if (!IS_ERR_OR_NULL(skb)) 1591 err = udp_v6_send_skb(skb, fl6, &cork.base); 1592 /* ip6_make_skb steals dst reference */ 1593 goto out_no_dst; 1594 } 1595 1596 lock_sock(sk); 1597 if (unlikely(up->pending)) { 1598 /* The socket is already corked while preparing it. */ 1599 /* ... which is an evident application bug. --ANK */ 1600 release_sock(sk); 1601 1602 net_dbg_ratelimited("udp cork app bug 2\n"); 1603 err = -EINVAL; 1604 goto out; 1605 } 1606 1607 WRITE_ONCE(up->pending, AF_INET6); 1608 1609 do_append_data: 1610 if (ipc6.dontfrag < 0) 1611 ipc6.dontfrag = inet6_test_bit(DONTFRAG, sk); 1612 up->len += ulen; 1613 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), 1614 &ipc6, fl6, (struct rt6_info *)dst, 1615 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 1616 if (err) 1617 udp_v6_flush_pending_frames(sk); 1618 else if (!corkreq) 1619 err = udp_v6_push_pending_frames(sk); 1620 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 1621 WRITE_ONCE(up->pending, 0); 1622 1623 if (err > 0) 1624 err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0; 1625 release_sock(sk); 1626 1627 out: 1628 dst_release(dst); 1629 out_no_dst: 1630 fl6_sock_release(flowlabel); 1631 txopt_put(opt_to_free); 1632 if (!err) 1633 return len; 1634 /* 1635 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 1636 * ENOBUFS might not be good (it's not tunable per se), but otherwise 1637 * we don't have a good statistic (IpOutDiscards but it can be too many 1638 * things). We could add another new stat but at least for now that 1639 * seems like overkill. 1640 */ 1641 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 1642 UDP6_INC_STATS(sock_net(sk), 1643 UDP_MIB_SNDBUFERRORS, is_udplite); 1644 } 1645 return err; 1646 1647 do_confirm: 1648 if (msg->msg_flags & MSG_PROBE) 1649 dst_confirm_neigh(dst, &fl6->daddr); 1650 if (!(msg->msg_flags&MSG_PROBE) || len) 1651 goto back_from_confirm; 1652 err = 0; 1653 goto out; 1654 } 1655 EXPORT_SYMBOL(udpv6_sendmsg); 1656 1657 static void udpv6_splice_eof(struct socket *sock) 1658 { 1659 struct sock *sk = sock->sk; 1660 struct udp_sock *up = udp_sk(sk); 1661 1662 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk)) 1663 return; 1664 1665 lock_sock(sk); 1666 if (up->pending && !udp_test_bit(CORK, sk)) 1667 udp_v6_push_pending_frames(sk); 1668 release_sock(sk); 1669 } 1670 1671 void udpv6_destroy_sock(struct sock *sk) 1672 { 1673 struct udp_sock *up = udp_sk(sk); 1674 lock_sock(sk); 1675 1676 /* protects from races with udp_abort() */ 1677 sock_set_flag(sk, SOCK_DEAD); 1678 udp_v6_flush_pending_frames(sk); 1679 release_sock(sk); 1680 1681 if (static_branch_unlikely(&udpv6_encap_needed_key)) { 1682 if (up->encap_type) { 1683 void (*encap_destroy)(struct sock *sk); 1684 encap_destroy = READ_ONCE(up->encap_destroy); 1685 if (encap_destroy) 1686 encap_destroy(sk); 1687 } 1688 if (udp_test_bit(ENCAP_ENABLED, sk)) { 1689 static_branch_dec(&udpv6_encap_needed_key); 1690 udp_encap_disable(); 1691 } 1692 } 1693 } 1694 1695 /* 1696 * Socket option code for UDP 1697 */ 1698 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval, 1699 unsigned int optlen) 1700 { 1701 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET) 1702 return udp_lib_setsockopt(sk, level, optname, 1703 optval, optlen, 1704 udp_v6_push_pending_frames); 1705 return ipv6_setsockopt(sk, level, optname, optval, optlen); 1706 } 1707 1708 int udpv6_getsockopt(struct sock *sk, int level, int optname, 1709 char __user *optval, int __user *optlen) 1710 { 1711 if (level == SOL_UDP || level == SOL_UDPLITE) 1712 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1713 return ipv6_getsockopt(sk, level, optname, optval, optlen); 1714 } 1715 1716 1717 /* ------------------------------------------------------------------------ */ 1718 #ifdef CONFIG_PROC_FS 1719 int udp6_seq_show(struct seq_file *seq, void *v) 1720 { 1721 if (v == SEQ_START_TOKEN) { 1722 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); 1723 } else { 1724 int bucket = ((struct udp_iter_state *)seq->private)->bucket; 1725 const struct inet_sock *inet = inet_sk((const struct sock *)v); 1726 __u16 srcp = ntohs(inet->inet_sport); 1727 __u16 destp = ntohs(inet->inet_dport); 1728 __ip6_dgram_sock_seq_show(seq, v, srcp, destp, 1729 udp_rqueue_get(v), bucket); 1730 } 1731 return 0; 1732 } 1733 1734 const struct seq_operations udp6_seq_ops = { 1735 .start = udp_seq_start, 1736 .next = udp_seq_next, 1737 .stop = udp_seq_stop, 1738 .show = udp6_seq_show, 1739 }; 1740 EXPORT_SYMBOL(udp6_seq_ops); 1741 1742 static struct udp_seq_afinfo udp6_seq_afinfo = { 1743 .family = AF_INET6, 1744 .udp_table = NULL, 1745 }; 1746 1747 int __net_init udp6_proc_init(struct net *net) 1748 { 1749 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops, 1750 sizeof(struct udp_iter_state), &udp6_seq_afinfo)) 1751 return -ENOMEM; 1752 return 0; 1753 } 1754 1755 void udp6_proc_exit(struct net *net) 1756 { 1757 remove_proc_entry("udp6", net->proc_net); 1758 } 1759 #endif /* CONFIG_PROC_FS */ 1760 1761 /* ------------------------------------------------------------------------ */ 1762 1763 struct proto udpv6_prot = { 1764 .name = "UDPv6", 1765 .owner = THIS_MODULE, 1766 .close = udp_lib_close, 1767 .pre_connect = udpv6_pre_connect, 1768 .connect = ip6_datagram_connect, 1769 .disconnect = udp_disconnect, 1770 .ioctl = udp_ioctl, 1771 .init = udpv6_init_sock, 1772 .destroy = udpv6_destroy_sock, 1773 .setsockopt = udpv6_setsockopt, 1774 .getsockopt = udpv6_getsockopt, 1775 .sendmsg = udpv6_sendmsg, 1776 .recvmsg = udpv6_recvmsg, 1777 .splice_eof = udpv6_splice_eof, 1778 .release_cb = ip6_datagram_release_cb, 1779 .hash = udp_lib_hash, 1780 .unhash = udp_lib_unhash, 1781 .rehash = udp_v6_rehash, 1782 .get_port = udp_v6_get_port, 1783 .put_port = udp_lib_unhash, 1784 #ifdef CONFIG_BPF_SYSCALL 1785 .psock_update_sk_prot = udp_bpf_update_proto, 1786 #endif 1787 1788 .memory_allocated = &udp_memory_allocated, 1789 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc, 1790 1791 .sysctl_mem = sysctl_udp_mem, 1792 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min), 1793 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min), 1794 .obj_size = sizeof(struct udp6_sock), 1795 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6), 1796 .h.udp_table = NULL, 1797 .diag_destroy = udp_abort, 1798 }; 1799 1800 static struct inet_protosw udpv6_protosw = { 1801 .type = SOCK_DGRAM, 1802 .protocol = IPPROTO_UDP, 1803 .prot = &udpv6_prot, 1804 .ops = &inet6_dgram_ops, 1805 .flags = INET_PROTOSW_PERMANENT, 1806 }; 1807 1808 int __init udpv6_init(void) 1809 { 1810 int ret; 1811 1812 net_hotdata.udpv6_protocol = (struct inet6_protocol) { 1813 .handler = udpv6_rcv, 1814 .err_handler = udpv6_err, 1815 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL, 1816 }; 1817 ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1818 if (ret) 1819 goto out; 1820 1821 ret = inet6_register_protosw(&udpv6_protosw); 1822 if (ret) 1823 goto out_udpv6_protocol; 1824 out: 1825 return ret; 1826 1827 out_udpv6_protocol: 1828 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1829 goto out; 1830 } 1831 1832 void udpv6_exit(void) 1833 { 1834 inet6_unregister_protosw(&udpv6_protosw); 1835 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP); 1836 } 1837